Additional reporting by Kevin Peachey and Dearbail Jordan
Boris Cherny, creator of Claude Code, still starts 80% of tasks in plan mode today. But with each new model generation, the one-shot success rate after planning keeps climbing. I think we're approaching the point where plan mode as a separate human-in-the-loop step fades away. Not because planning doesn't matter, but because models are getting good enough to plan well on their own. Big caveat: this only works if you've done the work in levels 3 through 6. If your context is clean, your constraints are explicit, your tools are well-described, and your feedback loops are tight, the model can plan reliably without you reviewing it first. If you haven't done that work, you'll still need to babysit the plan.。关于这个话题，新收录的资料提供了深入分析
。新收录的资料对此有专业解读
From end-position 43 to 46, we then see solid boosts in math scores (red = good, yay). But include layer 46 or beyond, and the benefits collapse again. The hypothesis: position 47 is where a different circuit begins. Including even one step of the next recipe messes up the current recipe.
本报北京3月9日电 （记者李龙伊）第十四届全国人大宪法和法律委员会9日下午召开全体会议，根据各代表团的审议意见，对生态环境法典草案、民族团结进步促进法草案、国家发展规划法草案进行统一审议。，更多细节参见新收录的资料
人 民 网 版 权 所 有 ，未 经 书 面 授 权 禁 止 使 用